2 * IPVS An implementation of the IP virtual server support for the
3 * LINUX operating system. IPVS is now implemented as a module
4 * over the Netfilter framework. IPVS can be used to build a
5 * high-performance and highly available server based on a
8 * Authors: Wensong Zhang <wensong@linuxvirtualserver.org>
9 * Peter Kese <peter.kese@ijs.si>
10 * Julian Anastasov <ja@ssi.bg>
12 * This program is free software; you can redistribute it and/or
13 * modify it under the terms of the GNU General Public License
14 * as published by the Free Software Foundation; either version
15 * 2 of the License, or (at your option) any later version.
17 * The IPVS code for kernel 2.2 was done by Wensong Zhang and Peter Kese,
18 * with changes/fixes from Julian Anastasov, Lars Marowsky-Bree, Horms
22 * Paul `Rusty' Russell properly handle non-linear skbs
23 * Harald Welte don't use nfcache
27 #define KMSG_COMPONENT "IPVS"
28 #define pr_fmt(fmt) KMSG_COMPONENT ": " fmt
30 #include <linux/module.h>
31 #include <linux/kernel.h>
33 #include <linux/tcp.h>
34 #include <linux/sctp.h>
35 #include <linux/icmp.h>
36 #include <linux/slab.h>
41 #include <net/icmp.h> /* for icmp_send */
42 #include <net/route.h>
43 #include <net/ip6_checksum.h>
44 #include <net/netns/generic.h> /* net_generic() */
46 #include <linux/netfilter.h>
47 #include <linux/netfilter_ipv4.h>
49 #ifdef CONFIG_IP_VS_IPV6
51 #include <linux/netfilter_ipv6.h>
52 #include <net/ip6_route.h>
55 #include <net/ip_vs.h>
58 EXPORT_SYMBOL(register_ip_vs_scheduler
);
59 EXPORT_SYMBOL(unregister_ip_vs_scheduler
);
60 EXPORT_SYMBOL(ip_vs_proto_name
);
61 EXPORT_SYMBOL(ip_vs_conn_new
);
62 EXPORT_SYMBOL(ip_vs_conn_in_get
);
63 EXPORT_SYMBOL(ip_vs_conn_out_get
);
64 #ifdef CONFIG_IP_VS_PROTO_TCP
65 EXPORT_SYMBOL(ip_vs_tcp_conn_listen
);
67 EXPORT_SYMBOL(ip_vs_conn_put
);
68 #ifdef CONFIG_IP_VS_DEBUG
69 EXPORT_SYMBOL(ip_vs_get_debug_level
);
71 EXPORT_SYMBOL(ip_vs_new_conn_out
);
73 static int ip_vs_net_id __read_mostly
;
74 /* netns cnt used for uniqueness */
75 static atomic_t ipvs_netns_cnt
= ATOMIC_INIT(0);
77 /* ID used in ICMP lookups */
78 #define icmp_id(icmph) (((icmph)->un).echo.id)
79 #define icmpv6_id(icmph) (icmph->icmp6_dataun.u_echo.identifier)
81 const char *ip_vs_proto_name(unsigned int proto
)
96 #ifdef CONFIG_IP_VS_IPV6
101 sprintf(buf
, "IP_%u", proto
);
106 void ip_vs_init_hash_table(struct list_head
*table
, int rows
)
109 INIT_LIST_HEAD(&table
[rows
]);
113 ip_vs_in_stats(struct ip_vs_conn
*cp
, struct sk_buff
*skb
)
115 struct ip_vs_dest
*dest
= cp
->dest
;
116 struct netns_ipvs
*ipvs
= cp
->ipvs
;
118 if (dest
&& (dest
->flags
& IP_VS_DEST_F_AVAILABLE
)) {
119 struct ip_vs_cpu_stats
*s
;
120 struct ip_vs_service
*svc
;
122 s
= this_cpu_ptr(dest
->stats
.cpustats
);
123 u64_stats_update_begin(&s
->syncp
);
125 s
->cnt
.inbytes
+= skb
->len
;
126 u64_stats_update_end(&s
->syncp
);
129 svc
= rcu_dereference(dest
->svc
);
130 s
= this_cpu_ptr(svc
->stats
.cpustats
);
131 u64_stats_update_begin(&s
->syncp
);
133 s
->cnt
.inbytes
+= skb
->len
;
134 u64_stats_update_end(&s
->syncp
);
137 s
= this_cpu_ptr(ipvs
->tot_stats
.cpustats
);
138 u64_stats_update_begin(&s
->syncp
);
140 s
->cnt
.inbytes
+= skb
->len
;
141 u64_stats_update_end(&s
->syncp
);
147 ip_vs_out_stats(struct ip_vs_conn
*cp
, struct sk_buff
*skb
)
149 struct ip_vs_dest
*dest
= cp
->dest
;
150 struct netns_ipvs
*ipvs
= cp
->ipvs
;
152 if (dest
&& (dest
->flags
& IP_VS_DEST_F_AVAILABLE
)) {
153 struct ip_vs_cpu_stats
*s
;
154 struct ip_vs_service
*svc
;
156 s
= this_cpu_ptr(dest
->stats
.cpustats
);
157 u64_stats_update_begin(&s
->syncp
);
159 s
->cnt
.outbytes
+= skb
->len
;
160 u64_stats_update_end(&s
->syncp
);
163 svc
= rcu_dereference(dest
->svc
);
164 s
= this_cpu_ptr(svc
->stats
.cpustats
);
165 u64_stats_update_begin(&s
->syncp
);
167 s
->cnt
.outbytes
+= skb
->len
;
168 u64_stats_update_end(&s
->syncp
);
171 s
= this_cpu_ptr(ipvs
->tot_stats
.cpustats
);
172 u64_stats_update_begin(&s
->syncp
);
174 s
->cnt
.outbytes
+= skb
->len
;
175 u64_stats_update_end(&s
->syncp
);
181 ip_vs_conn_stats(struct ip_vs_conn
*cp
, struct ip_vs_service
*svc
)
183 struct netns_ipvs
*ipvs
= svc
->ipvs
;
184 struct ip_vs_cpu_stats
*s
;
186 s
= this_cpu_ptr(cp
->dest
->stats
.cpustats
);
187 u64_stats_update_begin(&s
->syncp
);
189 u64_stats_update_end(&s
->syncp
);
191 s
= this_cpu_ptr(svc
->stats
.cpustats
);
192 u64_stats_update_begin(&s
->syncp
);
194 u64_stats_update_end(&s
->syncp
);
196 s
= this_cpu_ptr(ipvs
->tot_stats
.cpustats
);
197 u64_stats_update_begin(&s
->syncp
);
199 u64_stats_update_end(&s
->syncp
);
204 ip_vs_set_state(struct ip_vs_conn
*cp
, int direction
,
205 const struct sk_buff
*skb
,
206 struct ip_vs_proto_data
*pd
)
208 if (likely(pd
->pp
->state_transition
))
209 pd
->pp
->state_transition(cp
, direction
, skb
, pd
);
213 ip_vs_conn_fill_param_persist(const struct ip_vs_service
*svc
,
214 struct sk_buff
*skb
, int protocol
,
215 const union nf_inet_addr
*caddr
, __be16 cport
,
216 const union nf_inet_addr
*vaddr
, __be16 vport
,
217 struct ip_vs_conn_param
*p
)
219 ip_vs_conn_fill_param(svc
->ipvs
, svc
->af
, protocol
, caddr
, cport
, vaddr
,
221 p
->pe
= rcu_dereference(svc
->pe
);
222 if (p
->pe
&& p
->pe
->fill_param
)
223 return p
->pe
->fill_param(p
, skb
);
229 * IPVS persistent scheduling function
230 * It creates a connection entry according to its template if exists,
231 * or selects a server and creates a connection entry plus a template.
232 * Locking: we are svc user (svc->refcnt), so we hold all dests too
233 * Protocols supported: TCP, UDP
235 static struct ip_vs_conn
*
236 ip_vs_sched_persist(struct ip_vs_service
*svc
,
237 struct sk_buff
*skb
, __be16 src_port
, __be16 dst_port
,
238 int *ignored
, struct ip_vs_iphdr
*iph
)
240 struct ip_vs_conn
*cp
= NULL
;
241 struct ip_vs_dest
*dest
;
242 struct ip_vs_conn
*ct
;
243 __be16 dport
= 0; /* destination port to forward */
245 struct ip_vs_conn_param param
;
246 const union nf_inet_addr fwmark
= { .ip
= htonl(svc
->fwmark
) };
247 union nf_inet_addr snet
; /* source network of the client,
249 const union nf_inet_addr
*src_addr
, *dst_addr
;
251 if (likely(!ip_vs_iph_inverse(iph
))) {
252 src_addr
= &iph
->saddr
;
253 dst_addr
= &iph
->daddr
;
255 src_addr
= &iph
->daddr
;
256 dst_addr
= &iph
->saddr
;
260 /* Mask saddr with the netmask to adjust template granularity */
261 #ifdef CONFIG_IP_VS_IPV6
262 if (svc
->af
== AF_INET6
)
263 ipv6_addr_prefix(&snet
.in6
, &src_addr
->in6
,
264 (__force __u32
) svc
->netmask
);
267 snet
.ip
= src_addr
->ip
& svc
->netmask
;
269 IP_VS_DBG_BUF(6, "p-schedule: src %s:%u dest %s:%u "
271 IP_VS_DBG_ADDR(svc
->af
, src_addr
), ntohs(src_port
),
272 IP_VS_DBG_ADDR(svc
->af
, dst_addr
), ntohs(dst_port
),
273 IP_VS_DBG_ADDR(svc
->af
, &snet
));
276 * As far as we know, FTP is a very complicated network protocol, and
277 * it uses control connection and data connections. For active FTP,
278 * FTP server initialize data connection to the client, its source port
279 * is often 20. For passive FTP, FTP server tells the clients the port
280 * that it passively listens to, and the client issues the data
281 * connection. In the tunneling or direct routing mode, the load
282 * balancer is on the client-to-server half of connection, the port
283 * number is unknown to the load balancer. So, a conn template like
284 * <caddr, 0, vaddr, 0, daddr, 0> is created for persistent FTP
285 * service, and a template like <caddr, 0, vaddr, vport, daddr, dport>
286 * is created for other persistent services.
289 int protocol
= iph
->protocol
;
290 const union nf_inet_addr
*vaddr
= dst_addr
;
293 if (dst_port
== svc
->port
) {
295 * <protocol, caddr, 0, vaddr, vport, daddr, dport>
297 * <protocol, caddr, 0, vaddr, 0, daddr, 0>
299 if (svc
->port
!= FTPPORT
)
302 /* Note: persistent fwmark-based services and
303 * persistent port zero service are handled here.
305 * <IPPROTO_IP,caddr,0,fwmark,0,daddr,0>
306 * port zero template:
307 * <protocol,caddr,0,vaddr,0,daddr,0>
310 protocol
= IPPROTO_IP
;
314 /* return *ignored = -1 so NF_DROP can be used */
315 if (ip_vs_conn_fill_param_persist(svc
, skb
, protocol
, &snet
, 0,
316 vaddr
, vport
, ¶m
) < 0) {
322 /* Check if a template already exists */
323 ct
= ip_vs_ct_in_get(¶m
);
324 if (!ct
|| !ip_vs_check_template(ct
)) {
325 struct ip_vs_scheduler
*sched
;
328 * No template found or the dest of the connection
329 * template is not available.
330 * return *ignored=0 i.e. ICMP and NF_DROP
332 sched
= rcu_dereference(svc
->scheduler
);
334 /* read svc->sched_data after svc->scheduler */
336 dest
= sched
->schedule(svc
, skb
, iph
);
341 IP_VS_DBG(1, "p-schedule: no dest found.\n");
342 kfree(param
.pe_data
);
347 if (dst_port
== svc
->port
&& svc
->port
!= FTPPORT
)
351 * This adds param.pe_data to the template,
352 * and thus param.pe_data will be destroyed
353 * when the template expires */
354 ct
= ip_vs_conn_new(¶m
, dest
->af
, &dest
->addr
, dport
,
355 IP_VS_CONN_F_TEMPLATE
, dest
, skb
->mark
);
357 kfree(param
.pe_data
);
362 ct
->timeout
= svc
->timeout
;
364 /* set destination with the found template */
366 kfree(param
.pe_data
);
370 if (dport
== svc
->port
&& dest
->port
)
373 flags
= (svc
->flags
& IP_VS_SVC_F_ONEPACKET
374 && iph
->protocol
== IPPROTO_UDP
) ?
375 IP_VS_CONN_F_ONE_PACKET
: 0;
378 * Create a new connection according to the template
380 ip_vs_conn_fill_param(svc
->ipvs
, svc
->af
, iph
->protocol
, src_addr
,
381 src_port
, dst_addr
, dst_port
, ¶m
);
383 cp
= ip_vs_conn_new(¶m
, dest
->af
, &dest
->addr
, dport
, flags
, dest
,
394 ip_vs_control_add(cp
, ct
);
397 ip_vs_conn_stats(cp
, svc
);
403 * IPVS main scheduling function
404 * It selects a server according to the virtual service, and
405 * creates a connection entry.
406 * Protocols supported: TCP, UDP
410 * 1 : protocol tried to schedule (eg. on SYN), found svc but the
411 * svc/scheduler decides that this packet should be accepted with
412 * NF_ACCEPT because it must not be scheduled.
414 * 0 : scheduler can not find destination, so try bypass or
415 * return ICMP and then NF_DROP (ip_vs_leave).
417 * -1 : scheduler tried to schedule but fatal error occurred, eg.
418 * ip_vs_conn_new failure (ENOMEM) or ip_vs_sip_fill_param
419 * failure such as missing Call-ID, ENOMEM on skb_linearize
420 * or pe_data. In this case we should return NF_DROP without
421 * any attempts to send ICMP with ip_vs_leave.
424 ip_vs_schedule(struct ip_vs_service
*svc
, struct sk_buff
*skb
,
425 struct ip_vs_proto_data
*pd
, int *ignored
,
426 struct ip_vs_iphdr
*iph
)
428 struct ip_vs_protocol
*pp
= pd
->pp
;
429 struct ip_vs_conn
*cp
= NULL
;
430 struct ip_vs_scheduler
*sched
;
431 struct ip_vs_dest
*dest
;
432 __be16 _ports
[2], *pptr
, cport
, vport
;
433 const void *caddr
, *vaddr
;
438 * IPv6 frags, only the first hit here.
440 pptr
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_ports
), _ports
, iph
);
444 if (likely(!ip_vs_iph_inverse(iph
))) {
457 * FTPDATA needs this check when using local real server.
458 * Never schedule Active FTPDATA connections from real server.
459 * For LVS-NAT they must be already created. For other methods
460 * with persistence the connection is created on SYN+ACK.
462 if (cport
== FTPDATA
) {
463 IP_VS_DBG_PKT(12, svc
->af
, pp
, skb
, iph
->off
,
464 "Not scheduling FTPDATA");
469 * Do not schedule replies from local real server.
471 if ((!skb
->dev
|| skb
->dev
->flags
& IFF_LOOPBACK
)) {
472 iph
->hdr_flags
^= IP_VS_HDR_INVERSE
;
473 cp
= pp
->conn_in_get(svc
->ipvs
, svc
->af
, skb
, iph
);
474 iph
->hdr_flags
^= IP_VS_HDR_INVERSE
;
477 IP_VS_DBG_PKT(12, svc
->af
, pp
, skb
, iph
->off
,
478 "Not scheduling reply for existing"
480 __ip_vs_conn_put(cp
);
488 if (svc
->flags
& IP_VS_SVC_F_PERSISTENT
)
489 return ip_vs_sched_persist(svc
, skb
, cport
, vport
, ignored
,
495 * Non-persistent service
497 if (!svc
->fwmark
&& vport
!= svc
->port
) {
499 pr_err("Schedule: port zero only supported "
500 "in persistent services, "
501 "check your ipvs configuration\n");
505 sched
= rcu_dereference(svc
->scheduler
);
507 /* read svc->sched_data after svc->scheduler */
509 dest
= sched
->schedule(svc
, skb
, iph
);
514 IP_VS_DBG(1, "Schedule: no dest found.\n");
518 flags
= (svc
->flags
& IP_VS_SVC_F_ONEPACKET
519 && iph
->protocol
== IPPROTO_UDP
) ?
520 IP_VS_CONN_F_ONE_PACKET
: 0;
523 * Create a connection entry.
526 struct ip_vs_conn_param p
;
528 ip_vs_conn_fill_param(svc
->ipvs
, svc
->af
, iph
->protocol
,
529 caddr
, cport
, vaddr
, vport
, &p
);
530 cp
= ip_vs_conn_new(&p
, dest
->af
, &dest
->addr
,
531 dest
->port
? dest
->port
: vport
,
532 flags
, dest
, skb
->mark
);
539 IP_VS_DBG_BUF(6, "Schedule fwd:%c c:%s:%u v:%s:%u "
540 "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
542 IP_VS_DBG_ADDR(cp
->af
, &cp
->caddr
), ntohs(cp
->cport
),
543 IP_VS_DBG_ADDR(cp
->af
, &cp
->vaddr
), ntohs(cp
->vport
),
544 IP_VS_DBG_ADDR(cp
->daf
, &cp
->daddr
), ntohs(cp
->dport
),
545 cp
->flags
, atomic_read(&cp
->refcnt
));
547 ip_vs_conn_stats(cp
, svc
);
551 static inline int ip_vs_addr_is_unicast(struct net
*net
, int af
,
552 union nf_inet_addr
*addr
)
554 #ifdef CONFIG_IP_VS_IPV6
556 return ipv6_addr_type(&addr
->in6
) & IPV6_ADDR_UNICAST
;
558 return (inet_addr_type(net
, addr
->ip
) == RTN_UNICAST
);
562 * Pass or drop the packet.
563 * Called by ip_vs_in, when the virtual service is available but
564 * no destination is available for a new connection.
566 int ip_vs_leave(struct ip_vs_service
*svc
, struct sk_buff
*skb
,
567 struct ip_vs_proto_data
*pd
, struct ip_vs_iphdr
*iph
)
569 __be16 _ports
[2], *pptr
, dport
;
570 struct netns_ipvs
*ipvs
= svc
->ipvs
;
571 struct net
*net
= ipvs
->net
;
573 pptr
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_ports
), _ports
, iph
);
576 dport
= likely(!ip_vs_iph_inverse(iph
)) ? pptr
[1] : pptr
[0];
578 /* if it is fwmark-based service, the cache_bypass sysctl is up
579 and the destination is a non-local unicast, then create
580 a cache_bypass connection entry */
581 if (sysctl_cache_bypass(ipvs
) && svc
->fwmark
&&
582 !(iph
->hdr_flags
& (IP_VS_HDR_INVERSE
| IP_VS_HDR_ICMP
)) &&
583 ip_vs_addr_is_unicast(net
, svc
->af
, &iph
->daddr
)) {
585 struct ip_vs_conn
*cp
;
586 unsigned int flags
= (svc
->flags
& IP_VS_SVC_F_ONEPACKET
&&
587 iph
->protocol
== IPPROTO_UDP
) ?
588 IP_VS_CONN_F_ONE_PACKET
: 0;
589 union nf_inet_addr daddr
= { .all
= { 0, 0, 0, 0 } };
591 /* create a new connection entry */
592 IP_VS_DBG(6, "%s(): create a cache_bypass entry\n", __func__
);
594 struct ip_vs_conn_param p
;
595 ip_vs_conn_fill_param(svc
->ipvs
, svc
->af
, iph
->protocol
,
596 &iph
->saddr
, pptr
[0],
597 &iph
->daddr
, pptr
[1], &p
);
598 cp
= ip_vs_conn_new(&p
, svc
->af
, &daddr
, 0,
599 IP_VS_CONN_F_BYPASS
| flags
,
606 ip_vs_in_stats(cp
, skb
);
609 ip_vs_set_state(cp
, IP_VS_DIR_INPUT
, skb
, pd
);
611 /* transmit the first SYN packet */
612 ret
= cp
->packet_xmit(skb
, cp
, pd
->pp
, iph
);
613 /* do not touch skb anymore */
615 if ((cp
->flags
& IP_VS_CONN_F_ONE_PACKET
) && cp
->control
)
616 atomic_inc(&cp
->control
->in_pkts
);
618 atomic_inc(&cp
->in_pkts
);
624 * When the virtual ftp service is presented, packets destined
625 * for other services on the VIP may get here (except services
626 * listed in the ipvs table), pass the packets, because it is
627 * not ipvs job to decide to drop the packets.
629 if (svc
->port
== FTPPORT
&& dport
!= FTPPORT
)
632 if (unlikely(ip_vs_iph_icmp(iph
)))
636 * Notify the client that the destination is unreachable, and
637 * release the socket buffer.
638 * Since it is in IP layer, the TCP socket is not actually
639 * created, the TCP RST packet cannot be sent, instead that
640 * ICMP_PORT_UNREACH is sent here no matter it is TCP/UDP. --WZ
642 #ifdef CONFIG_IP_VS_IPV6
643 if (svc
->af
== AF_INET6
) {
645 skb
->dev
= net
->loopback_dev
;
646 icmpv6_send(skb
, ICMPV6_DEST_UNREACH
, ICMPV6_PORT_UNREACH
, 0);
649 icmp_send(skb
, ICMP_DEST_UNREACH
, ICMP_PORT_UNREACH
, 0);
656 static int sysctl_snat_reroute(struct netns_ipvs
*ipvs
)
658 return ipvs
->sysctl_snat_reroute
;
661 static int sysctl_nat_icmp_send(struct netns_ipvs
*ipvs
)
663 return ipvs
->sysctl_nat_icmp_send
;
666 static int sysctl_expire_nodest_conn(struct netns_ipvs
*ipvs
)
668 return ipvs
->sysctl_expire_nodest_conn
;
673 static int sysctl_snat_reroute(struct netns_ipvs
*ipvs
) { return 0; }
674 static int sysctl_nat_icmp_send(struct netns_ipvs
*ipvs
) { return 0; }
675 static int sysctl_expire_nodest_conn(struct netns_ipvs
*ipvs
) { return 0; }
679 __sum16
ip_vs_checksum_complete(struct sk_buff
*skb
, int offset
)
681 return csum_fold(skb_checksum(skb
, offset
, skb
->len
- offset
, 0));
684 static inline enum ip_defrag_users
ip_vs_defrag_user(unsigned int hooknum
)
686 if (NF_INET_LOCAL_IN
== hooknum
)
687 return IP_DEFRAG_VS_IN
;
688 if (NF_INET_FORWARD
== hooknum
)
689 return IP_DEFRAG_VS_FWD
;
690 return IP_DEFRAG_VS_OUT
;
693 static inline int ip_vs_gather_frags(struct netns_ipvs
*ipvs
,
694 struct sk_buff
*skb
, u_int32_t user
)
699 err
= ip_defrag(ipvs
->net
, skb
, user
);
702 ip_send_check(ip_hdr(skb
));
707 static int ip_vs_route_me_harder(struct netns_ipvs
*ipvs
, int af
,
708 struct sk_buff
*skb
, unsigned int hooknum
)
710 if (!sysctl_snat_reroute(ipvs
))
712 /* Reroute replies only to remote clients (FORWARD and LOCAL_OUT) */
713 if (NF_INET_LOCAL_IN
== hooknum
)
715 #ifdef CONFIG_IP_VS_IPV6
716 if (af
== AF_INET6
) {
717 struct dst_entry
*dst
= skb_dst(skb
);
719 if (dst
->dev
&& !(dst
->dev
->flags
& IFF_LOOPBACK
) &&
720 ip6_route_me_harder(ipvs
->net
, skb
) != 0)
724 if (!(skb_rtable(skb
)->rt_flags
& RTCF_LOCAL
) &&
725 ip_route_me_harder(ipvs
->net
, skb
, RTN_LOCAL
) != 0)
732 * Packet has been made sufficiently writable in caller
733 * - inout: 1=in->out, 0=out->in
735 void ip_vs_nat_icmp(struct sk_buff
*skb
, struct ip_vs_protocol
*pp
,
736 struct ip_vs_conn
*cp
, int inout
)
738 struct iphdr
*iph
= ip_hdr(skb
);
739 unsigned int icmp_offset
= iph
->ihl
*4;
740 struct icmphdr
*icmph
= (struct icmphdr
*)(skb_network_header(skb
) +
742 struct iphdr
*ciph
= (struct iphdr
*)(icmph
+ 1);
745 iph
->saddr
= cp
->vaddr
.ip
;
747 ciph
->daddr
= cp
->vaddr
.ip
;
750 iph
->daddr
= cp
->daddr
.ip
;
752 ciph
->saddr
= cp
->daddr
.ip
;
756 /* the TCP/UDP/SCTP port */
757 if (IPPROTO_TCP
== ciph
->protocol
|| IPPROTO_UDP
== ciph
->protocol
||
758 IPPROTO_SCTP
== ciph
->protocol
) {
759 __be16
*ports
= (void *)ciph
+ ciph
->ihl
*4;
762 ports
[1] = cp
->vport
;
764 ports
[0] = cp
->dport
;
767 /* And finally the ICMP checksum */
769 icmph
->checksum
= ip_vs_checksum_complete(skb
, icmp_offset
);
770 skb
->ip_summed
= CHECKSUM_UNNECESSARY
;
773 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, (void *)ciph
- (void *)iph
,
774 "Forwarding altered outgoing ICMP");
776 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, (void *)ciph
- (void *)iph
,
777 "Forwarding altered incoming ICMP");
780 #ifdef CONFIG_IP_VS_IPV6
781 void ip_vs_nat_icmp_v6(struct sk_buff
*skb
, struct ip_vs_protocol
*pp
,
782 struct ip_vs_conn
*cp
, int inout
)
784 struct ipv6hdr
*iph
= ipv6_hdr(skb
);
785 unsigned int icmp_offset
= 0;
786 unsigned int offs
= 0; /* header offset*/
788 struct icmp6hdr
*icmph
;
789 struct ipv6hdr
*ciph
;
790 unsigned short fragoffs
;
792 ipv6_find_hdr(skb
, &icmp_offset
, IPPROTO_ICMPV6
, &fragoffs
, NULL
);
793 icmph
= (struct icmp6hdr
*)(skb_network_header(skb
) + icmp_offset
);
794 offs
= icmp_offset
+ sizeof(struct icmp6hdr
);
795 ciph
= (struct ipv6hdr
*)(skb_network_header(skb
) + offs
);
797 protocol
= ipv6_find_hdr(skb
, &offs
, -1, &fragoffs
, NULL
);
800 iph
->saddr
= cp
->vaddr
.in6
;
801 ciph
->daddr
= cp
->vaddr
.in6
;
803 iph
->daddr
= cp
->daddr
.in6
;
804 ciph
->saddr
= cp
->daddr
.in6
;
807 /* the TCP/UDP/SCTP port */
808 if (!fragoffs
&& (IPPROTO_TCP
== protocol
|| IPPROTO_UDP
== protocol
||
809 IPPROTO_SCTP
== protocol
)) {
810 __be16
*ports
= (void *)(skb_network_header(skb
) + offs
);
812 IP_VS_DBG(11, "%s() changed port %d to %d\n", __func__
,
813 ntohs(inout
? ports
[1] : ports
[0]),
814 ntohs(inout
? cp
->vport
: cp
->dport
));
816 ports
[1] = cp
->vport
;
818 ports
[0] = cp
->dport
;
821 /* And finally the ICMP checksum */
822 icmph
->icmp6_cksum
= ~csum_ipv6_magic(&iph
->saddr
, &iph
->daddr
,
823 skb
->len
- icmp_offset
,
825 skb
->csum_start
= skb_network_header(skb
) - skb
->head
+ icmp_offset
;
826 skb
->csum_offset
= offsetof(struct icmp6hdr
, icmp6_cksum
);
827 skb
->ip_summed
= CHECKSUM_PARTIAL
;
830 IP_VS_DBG_PKT(11, AF_INET6
, pp
, skb
,
831 (void *)ciph
- (void *)iph
,
832 "Forwarding altered outgoing ICMPv6");
834 IP_VS_DBG_PKT(11, AF_INET6
, pp
, skb
,
835 (void *)ciph
- (void *)iph
,
836 "Forwarding altered incoming ICMPv6");
840 /* Handle relevant response ICMP messages - forward to the right
843 static int handle_response_icmp(int af
, struct sk_buff
*skb
,
844 union nf_inet_addr
*snet
,
845 __u8 protocol
, struct ip_vs_conn
*cp
,
846 struct ip_vs_protocol
*pp
,
847 unsigned int offset
, unsigned int ihl
,
848 unsigned int hooknum
)
850 unsigned int verdict
= NF_DROP
;
852 if (IP_VS_FWD_METHOD(cp
) != 0) {
853 pr_err("shouldn't reach here, because the box is on the "
854 "half connection in the tun/dr module.\n");
857 /* Ensure the checksum is correct */
858 if (!skb_csum_unnecessary(skb
) && ip_vs_checksum_complete(skb
, ihl
)) {
859 /* Failed checksum! */
860 IP_VS_DBG_BUF(1, "Forward ICMP: failed checksum from %s!\n",
861 IP_VS_DBG_ADDR(af
, snet
));
865 if (IPPROTO_TCP
== protocol
|| IPPROTO_UDP
== protocol
||
866 IPPROTO_SCTP
== protocol
)
867 offset
+= 2 * sizeof(__u16
);
868 if (!skb_make_writable(skb
, offset
))
871 #ifdef CONFIG_IP_VS_IPV6
873 ip_vs_nat_icmp_v6(skb
, pp
, cp
, 1);
876 ip_vs_nat_icmp(skb
, pp
, cp
, 1);
878 if (ip_vs_route_me_harder(cp
->ipvs
, af
, skb
, hooknum
))
881 /* do the statistics and put it back */
882 ip_vs_out_stats(cp
, skb
);
884 skb
->ipvs_property
= 1;
885 if (!(cp
->flags
& IP_VS_CONN_F_NFCT
))
888 ip_vs_update_conntrack(skb
, cp
, 0);
892 __ip_vs_conn_put(cp
);
898 * Handle ICMP messages in the inside-to-outside direction (outgoing).
899 * Find any that might be relevant, check against existing connections.
900 * Currently handles error types - unreachable, quench, ttl exceeded.
902 static int ip_vs_out_icmp(struct netns_ipvs
*ipvs
, struct sk_buff
*skb
,
903 int *related
, unsigned int hooknum
)
906 struct icmphdr _icmph
, *ic
;
907 struct iphdr _ciph
, *cih
; /* The ip header contained within the ICMP */
908 struct ip_vs_iphdr ciph
;
909 struct ip_vs_conn
*cp
;
910 struct ip_vs_protocol
*pp
;
911 unsigned int offset
, ihl
;
912 union nf_inet_addr snet
;
916 /* reassemble IP fragments */
917 if (ip_is_fragment(ip_hdr(skb
))) {
918 if (ip_vs_gather_frags(ipvs
, skb
, ip_vs_defrag_user(hooknum
)))
923 offset
= ihl
= iph
->ihl
* 4;
924 ic
= skb_header_pointer(skb
, offset
, sizeof(_icmph
), &_icmph
);
928 IP_VS_DBG(12, "Outgoing ICMP (%d,%d) %pI4->%pI4\n",
929 ic
->type
, ntohs(icmp_id(ic
)),
930 &iph
->saddr
, &iph
->daddr
);
933 * Work through seeing if this is for us.
934 * These checks are supposed to be in an order that means easy
935 * things are checked first to speed up processing.... however
936 * this means that some packets will manage to get a long way
937 * down this stack and then be rejected, but that's life.
939 if ((ic
->type
!= ICMP_DEST_UNREACH
) &&
940 (ic
->type
!= ICMP_SOURCE_QUENCH
) &&
941 (ic
->type
!= ICMP_TIME_EXCEEDED
)) {
946 /* Now find the contained IP header */
947 offset
+= sizeof(_icmph
);
948 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
950 return NF_ACCEPT
; /* The packet looks wrong, ignore */
952 pp
= ip_vs_proto_get(cih
->protocol
);
956 /* Is the embedded protocol header present? */
957 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
) &&
961 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, offset
,
962 "Checking outgoing ICMP for");
964 ip_vs_fill_iph_skb_icmp(AF_INET
, skb
, offset
, true, &ciph
);
966 /* The embedded headers contain source and dest in reverse order */
967 cp
= pp
->conn_out_get(ipvs
, AF_INET
, skb
, &ciph
);
971 snet
.ip
= iph
->saddr
;
972 return handle_response_icmp(AF_INET
, skb
, &snet
, cih
->protocol
, cp
,
973 pp
, ciph
.len
, ihl
, hooknum
);
976 #ifdef CONFIG_IP_VS_IPV6
977 static int ip_vs_out_icmp_v6(struct netns_ipvs
*ipvs
, struct sk_buff
*skb
,
978 int *related
, unsigned int hooknum
,
979 struct ip_vs_iphdr
*ipvsh
)
981 struct icmp6hdr _icmph
, *ic
;
982 struct ip_vs_iphdr ciph
= {.flags
= 0, .fragoffs
= 0};/*Contained IP */
983 struct ip_vs_conn
*cp
;
984 struct ip_vs_protocol
*pp
;
985 union nf_inet_addr snet
;
989 ic
= frag_safe_skb_hp(skb
, ipvsh
->len
, sizeof(_icmph
), &_icmph
, ipvsh
);
994 * Work through seeing if this is for us.
995 * These checks are supposed to be in an order that means easy
996 * things are checked first to speed up processing.... however
997 * this means that some packets will manage to get a long way
998 * down this stack and then be rejected, but that's life.
1000 if (ic
->icmp6_type
& ICMPV6_INFOMSG_MASK
) {
1004 /* Fragment header that is before ICMP header tells us that:
1005 * it's not an error message since they can't be fragmented.
1007 if (ipvsh
->flags
& IP6_FH_F_FRAG
)
1010 IP_VS_DBG(8, "Outgoing ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1011 ic
->icmp6_type
, ntohs(icmpv6_id(ic
)),
1012 &ipvsh
->saddr
, &ipvsh
->daddr
);
1014 if (!ip_vs_fill_iph_skb_icmp(AF_INET6
, skb
, ipvsh
->len
+ sizeof(_icmph
),
1016 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1018 pp
= ip_vs_proto_get(ciph
.protocol
);
1022 /* The embedded headers contain source and dest in reverse order */
1023 cp
= pp
->conn_out_get(ipvs
, AF_INET6
, skb
, &ciph
);
1027 snet
.in6
= ciph
.saddr
.in6
;
1029 return handle_response_icmp(AF_INET6
, skb
, &snet
, ciph
.protocol
, cp
,
1030 pp
, offset
, sizeof(struct ipv6hdr
),
1036 * Check if sctp chunc is ABORT chunk
1038 static inline int is_sctp_abort(const struct sk_buff
*skb
, int nh_len
)
1040 sctp_chunkhdr_t
*sch
, schunk
;
1041 sch
= skb_header_pointer(skb
, nh_len
+ sizeof(sctp_sctphdr_t
),
1042 sizeof(schunk
), &schunk
);
1045 if (sch
->type
== SCTP_CID_ABORT
)
1050 static inline int is_tcp_reset(const struct sk_buff
*skb
, int nh_len
)
1052 struct tcphdr _tcph
, *th
;
1054 th
= skb_header_pointer(skb
, nh_len
, sizeof(_tcph
), &_tcph
);
1060 static inline bool is_new_conn(const struct sk_buff
*skb
,
1061 struct ip_vs_iphdr
*iph
)
1063 switch (iph
->protocol
) {
1065 struct tcphdr _tcph
, *th
;
1067 th
= skb_header_pointer(skb
, iph
->len
, sizeof(_tcph
), &_tcph
);
1072 case IPPROTO_SCTP
: {
1073 sctp_chunkhdr_t
*sch
, schunk
;
1075 sch
= skb_header_pointer(skb
, iph
->len
+ sizeof(sctp_sctphdr_t
),
1076 sizeof(schunk
), &schunk
);
1079 return sch
->type
== SCTP_CID_INIT
;
1086 static inline bool is_new_conn_expected(const struct ip_vs_conn
*cp
,
1087 int conn_reuse_mode
)
1089 /* Controlled (FTP DATA or persistence)? */
1093 switch (cp
->protocol
) {
1095 return (cp
->state
== IP_VS_TCP_S_TIME_WAIT
) ||
1096 (cp
->state
== IP_VS_TCP_S_CLOSE
) ||
1097 ((conn_reuse_mode
& 2) &&
1098 (cp
->state
== IP_VS_TCP_S_FIN_WAIT
) &&
1099 (cp
->flags
& IP_VS_CONN_F_NOOUTPUT
));
1101 return cp
->state
== IP_VS_SCTP_S_CLOSED
;
1107 /* Generic function to create new connections for outgoing RS packets
1109 * Pre-requisites for successful connection creation:
1110 * 1) Virtual Service is NOT fwmark based:
1111 * In fwmark-VS actual vaddr and vport are unknown to IPVS
1112 * 2) Real Server and Virtual Service were NOT configured without port:
1113 * This is to allow match of different VS to the same RS ip-addr
1115 struct ip_vs_conn
*ip_vs_new_conn_out(struct ip_vs_service
*svc
,
1116 struct ip_vs_dest
*dest
,
1117 struct sk_buff
*skb
,
1118 const struct ip_vs_iphdr
*iph
,
1122 struct ip_vs_conn_param param
;
1123 struct ip_vs_conn
*ct
= NULL
, *cp
= NULL
;
1124 const union nf_inet_addr
*vaddr
, *daddr
, *caddr
;
1125 union nf_inet_addr snet
;
1132 daddr
= &iph
->saddr
;
1133 caddr
= &iph
->daddr
;
1135 /* check pre-requisites are satisfied */
1138 if (!vport
|| !dport
)
1141 /* for persistent service first create connection template */
1142 if (svc
->flags
& IP_VS_SVC_F_PERSISTENT
) {
1143 /* apply netmask the same way ingress-side does */
1144 #ifdef CONFIG_IP_VS_IPV6
1145 if (svc
->af
== AF_INET6
)
1146 ipv6_addr_prefix(&snet
.in6
, &caddr
->in6
,
1147 (__force __u32
)svc
->netmask
);
1150 snet
.ip
= caddr
->ip
& svc
->netmask
;
1151 /* fill params and create template if not existent */
1152 if (ip_vs_conn_fill_param_persist(svc
, skb
, iph
->protocol
,
1156 ct
= ip_vs_ct_in_get(¶m
);
1158 ct
= ip_vs_conn_new(¶m
, dest
->af
, daddr
, dport
,
1159 IP_VS_CONN_F_TEMPLATE
, dest
, 0);
1161 kfree(param
.pe_data
);
1164 ct
->timeout
= svc
->timeout
;
1166 kfree(param
.pe_data
);
1170 /* connection flags */
1171 flags
= ((svc
->flags
& IP_VS_SVC_F_ONEPACKET
) &&
1172 iph
->protocol
== IPPROTO_UDP
) ? IP_VS_CONN_F_ONE_PACKET
: 0;
1173 /* create connection */
1174 ip_vs_conn_fill_param(svc
->ipvs
, svc
->af
, iph
->protocol
,
1175 caddr
, cport
, vaddr
, vport
, ¶m
);
1176 cp
= ip_vs_conn_new(¶m
, dest
->af
, daddr
, dport
, flags
, dest
, 0);
1183 ip_vs_control_add(cp
, ct
);
1186 ip_vs_conn_stats(cp
, svc
);
1188 /* return connection (will be used to handle outgoing packet) */
1189 IP_VS_DBG_BUF(6, "New connection RS-initiated:%c c:%s:%u v:%s:%u "
1190 "d:%s:%u conn->flags:%X conn->refcnt:%d\n",
1192 IP_VS_DBG_ADDR(cp
->af
, &cp
->caddr
), ntohs(cp
->cport
),
1193 IP_VS_DBG_ADDR(cp
->af
, &cp
->vaddr
), ntohs(cp
->vport
),
1194 IP_VS_DBG_ADDR(cp
->af
, &cp
->daddr
), ntohs(cp
->dport
),
1195 cp
->flags
, atomic_read(&cp
->refcnt
));
1200 /* Handle outgoing packets which are considered requests initiated by
1201 * real servers, so that subsequent responses from external client can be
1202 * routed to the right real server.
1203 * Used also for outgoing responses in OPS mode.
1205 * Connection management is handled by persistent-engine specific callback.
1207 static struct ip_vs_conn
*__ip_vs_rs_conn_out(unsigned int hooknum
,
1208 struct netns_ipvs
*ipvs
,
1209 int af
, struct sk_buff
*skb
,
1210 const struct ip_vs_iphdr
*iph
)
1212 struct ip_vs_dest
*dest
;
1213 struct ip_vs_conn
*cp
= NULL
;
1214 __be16 _ports
[2], *pptr
;
1216 if (hooknum
== NF_INET_LOCAL_IN
)
1219 pptr
= frag_safe_skb_hp(skb
, iph
->len
,
1220 sizeof(_ports
), _ports
, iph
);
1225 dest
= ip_vs_find_real_service(ipvs
, af
, iph
->protocol
,
1226 &iph
->saddr
, pptr
[0]);
1228 struct ip_vs_service
*svc
;
1229 struct ip_vs_pe
*pe
;
1231 svc
= rcu_dereference(dest
->svc
);
1233 pe
= rcu_dereference(svc
->pe
);
1234 if (pe
&& pe
->conn_out
)
1235 cp
= pe
->conn_out(svc
, dest
, skb
, iph
,
1244 /* Handle response packets: rewrite addresses and send away...
1247 handle_response(int af
, struct sk_buff
*skb
, struct ip_vs_proto_data
*pd
,
1248 struct ip_vs_conn
*cp
, struct ip_vs_iphdr
*iph
,
1249 unsigned int hooknum
)
1251 struct ip_vs_protocol
*pp
= pd
->pp
;
1253 IP_VS_DBG_PKT(11, af
, pp
, skb
, iph
->off
, "Outgoing packet");
1255 if (!skb_make_writable(skb
, iph
->len
))
1258 /* mangle the packet */
1259 if (pp
->snat_handler
&& !pp
->snat_handler(skb
, pp
, cp
, iph
))
1262 #ifdef CONFIG_IP_VS_IPV6
1264 ipv6_hdr(skb
)->saddr
= cp
->vaddr
.in6
;
1268 ip_hdr(skb
)->saddr
= cp
->vaddr
.ip
;
1269 ip_send_check(ip_hdr(skb
));
1273 * nf_iterate does not expect change in the skb->dst->dev.
1274 * It looks like it is not fatal to enable this code for hooks
1275 * where our handlers are at the end of the chain list and
1276 * when all next handlers use skb->dst->dev and not outdev.
1277 * It will definitely route properly the inout NAT traffic
1278 * when multiple paths are used.
1281 /* For policy routing, packets originating from this
1282 * machine itself may be routed differently to packets
1283 * passing through. We want this packet to be routed as
1284 * if it came from this machine itself. So re-compute
1285 * the routing information.
1287 if (ip_vs_route_me_harder(cp
->ipvs
, af
, skb
, hooknum
))
1290 IP_VS_DBG_PKT(10, af
, pp
, skb
, iph
->off
, "After SNAT");
1292 ip_vs_out_stats(cp
, skb
);
1293 ip_vs_set_state(cp
, IP_VS_DIR_OUTPUT
, skb
, pd
);
1294 skb
->ipvs_property
= 1;
1295 if (!(cp
->flags
& IP_VS_CONN_F_NFCT
))
1298 ip_vs_update_conntrack(skb
, cp
, 0);
1312 * Check if outgoing packet belongs to the established ip_vs_conn.
1315 ip_vs_out(struct netns_ipvs
*ipvs
, unsigned int hooknum
, struct sk_buff
*skb
, int af
)
1317 struct ip_vs_iphdr iph
;
1318 struct ip_vs_protocol
*pp
;
1319 struct ip_vs_proto_data
*pd
;
1320 struct ip_vs_conn
*cp
;
1325 /* Already marked as IPVS request or reply? */
1326 if (skb
->ipvs_property
)
1329 sk
= skb_to_full_sk(skb
);
1330 /* Bad... Do not break raw sockets */
1331 if (unlikely(sk
&& hooknum
== NF_INET_LOCAL_OUT
&&
1334 if (sk
->sk_family
== PF_INET
&& inet_sk(sk
)->nodefrag
)
1338 if (unlikely(!skb_dst(skb
)))
1344 ip_vs_fill_iph_skb(af
, skb
, false, &iph
);
1345 #ifdef CONFIG_IP_VS_IPV6
1346 if (af
== AF_INET6
) {
1347 if (unlikely(iph
.protocol
== IPPROTO_ICMPV6
)) {
1349 int verdict
= ip_vs_out_icmp_v6(ipvs
, skb
, &related
,
1357 if (unlikely(iph
.protocol
== IPPROTO_ICMP
)) {
1359 int verdict
= ip_vs_out_icmp(ipvs
, skb
, &related
, hooknum
);
1365 pd
= ip_vs_proto_data_get(ipvs
, iph
.protocol
);
1370 /* reassemble IP fragments */
1371 #ifdef CONFIG_IP_VS_IPV6
1374 if (unlikely(ip_is_fragment(ip_hdr(skb
)) && !pp
->dont_defrag
)) {
1375 if (ip_vs_gather_frags(ipvs
, skb
,
1376 ip_vs_defrag_user(hooknum
)))
1379 ip_vs_fill_iph_skb(AF_INET
, skb
, false, &iph
);
1383 * Check if the packet belongs to an existing entry
1385 cp
= pp
->conn_out_get(ipvs
, af
, skb
, &iph
);
1388 return handle_response(af
, skb
, pd
, cp
, &iph
, hooknum
);
1390 /* Check for real-server-started requests */
1391 if (atomic_read(&ipvs
->conn_out_counter
)) {
1392 /* Currently only for UDP:
1393 * connection oriented protocols typically use
1394 * ephemeral ports for outgoing connections, so
1395 * related incoming responses would not match any VS
1397 if (pp
->protocol
== IPPROTO_UDP
) {
1398 cp
= __ip_vs_rs_conn_out(hooknum
, ipvs
, af
, skb
, &iph
);
1400 return handle_response(af
, skb
, pd
, cp
, &iph
,
1405 if (sysctl_nat_icmp_send(ipvs
) &&
1406 (pp
->protocol
== IPPROTO_TCP
||
1407 pp
->protocol
== IPPROTO_UDP
||
1408 pp
->protocol
== IPPROTO_SCTP
)) {
1409 __be16 _ports
[2], *pptr
;
1411 pptr
= frag_safe_skb_hp(skb
, iph
.len
,
1412 sizeof(_ports
), _ports
, &iph
);
1414 return NF_ACCEPT
; /* Not for me */
1415 if (ip_vs_has_real_service(ipvs
, af
, iph
.protocol
, &iph
.saddr
,
1418 * Notify the real server: there is no
1419 * existing entry if it is not RST
1420 * packet or not TCP packet.
1422 if ((iph
.protocol
!= IPPROTO_TCP
&&
1423 iph
.protocol
!= IPPROTO_SCTP
)
1424 || ((iph
.protocol
== IPPROTO_TCP
1425 && !is_tcp_reset(skb
, iph
.len
))
1426 || (iph
.protocol
== IPPROTO_SCTP
1427 && !is_sctp_abort(skb
,
1429 #ifdef CONFIG_IP_VS_IPV6
1430 if (af
== AF_INET6
) {
1432 skb
->dev
= ipvs
->net
->loopback_dev
;
1434 ICMPV6_DEST_UNREACH
,
1435 ICMPV6_PORT_UNREACH
,
1441 ICMP_PORT_UNREACH
, 0);
1446 IP_VS_DBG_PKT(12, af
, pp
, skb
, iph
.off
,
1447 "ip_vs_out: packet continues traversal as normal");
1452 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1453 * used only for VS/NAT.
1454 * Check if packet is reply for established ip_vs_conn.
1457 ip_vs_reply4(void *priv
, struct sk_buff
*skb
,
1458 const struct nf_hook_state
*state
)
1460 return ip_vs_out(net_ipvs(state
->net
), state
->hook
, skb
, AF_INET
);
1464 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1465 * Check if packet is reply for established ip_vs_conn.
1468 ip_vs_local_reply4(void *priv
, struct sk_buff
*skb
,
1469 const struct nf_hook_state
*state
)
1471 return ip_vs_out(net_ipvs(state
->net
), state
->hook
, skb
, AF_INET
);
1474 #ifdef CONFIG_IP_VS_IPV6
1477 * It is hooked at the NF_INET_FORWARD and NF_INET_LOCAL_IN chain,
1478 * used only for VS/NAT.
1479 * Check if packet is reply for established ip_vs_conn.
1482 ip_vs_reply6(void *priv
, struct sk_buff
*skb
,
1483 const struct nf_hook_state
*state
)
1485 return ip_vs_out(net_ipvs(state
->net
), state
->hook
, skb
, AF_INET6
);
1489 * It is hooked at the NF_INET_LOCAL_OUT chain, used only for VS/NAT.
1490 * Check if packet is reply for established ip_vs_conn.
1493 ip_vs_local_reply6(void *priv
, struct sk_buff
*skb
,
1494 const struct nf_hook_state
*state
)
1496 return ip_vs_out(net_ipvs(state
->net
), state
->hook
, skb
, AF_INET6
);
1502 ip_vs_try_to_schedule(struct netns_ipvs
*ipvs
, int af
, struct sk_buff
*skb
,
1503 struct ip_vs_proto_data
*pd
,
1504 int *verdict
, struct ip_vs_conn
**cpp
,
1505 struct ip_vs_iphdr
*iph
)
1507 struct ip_vs_protocol
*pp
= pd
->pp
;
1509 if (!iph
->fragoffs
) {
1510 /* No (second) fragments need to enter here, as nf_defrag_ipv6
1511 * replayed fragment zero will already have created the cp
1514 /* Schedule and create new connection entry into cpp */
1515 if (!pp
->conn_schedule(ipvs
, af
, skb
, pd
, verdict
, cpp
, iph
))
1519 if (unlikely(!*cpp
)) {
1520 /* sorry, all this trouble for a no-hit :) */
1521 IP_VS_DBG_PKT(12, af
, pp
, skb
, iph
->off
,
1522 "ip_vs_in: packet continues traversal as normal");
1523 if (iph
->fragoffs
) {
1524 /* Fragment that couldn't be mapped to a conn entry
1525 * is missing module nf_defrag_ipv6
1527 IP_VS_DBG_RL("Unhandled frag, load nf_defrag_ipv6\n");
1528 IP_VS_DBG_PKT(7, af
, pp
, skb
, iph
->off
,
1529 "unhandled fragment");
1531 *verdict
= NF_ACCEPT
;
1539 * Handle ICMP messages in the outside-to-inside direction (incoming).
1540 * Find any that might be relevant, check against existing connections,
1541 * forward to the right destination host if relevant.
1542 * Currently handles error types - unreachable, quench, ttl exceeded.
1545 ip_vs_in_icmp(struct netns_ipvs
*ipvs
, struct sk_buff
*skb
, int *related
,
1546 unsigned int hooknum
)
1549 struct icmphdr _icmph
, *ic
;
1550 struct iphdr _ciph
, *cih
; /* The ip header contained within the ICMP */
1551 struct ip_vs_iphdr ciph
;
1552 struct ip_vs_conn
*cp
;
1553 struct ip_vs_protocol
*pp
;
1554 struct ip_vs_proto_data
*pd
;
1555 unsigned int offset
, offset2
, ihl
, verdict
;
1556 bool ipip
, new_cp
= false;
1560 /* reassemble IP fragments */
1561 if (ip_is_fragment(ip_hdr(skb
))) {
1562 if (ip_vs_gather_frags(ipvs
, skb
, ip_vs_defrag_user(hooknum
)))
1567 offset
= ihl
= iph
->ihl
* 4;
1568 ic
= skb_header_pointer(skb
, offset
, sizeof(_icmph
), &_icmph
);
1572 IP_VS_DBG(12, "Incoming ICMP (%d,%d) %pI4->%pI4\n",
1573 ic
->type
, ntohs(icmp_id(ic
)),
1574 &iph
->saddr
, &iph
->daddr
);
1577 * Work through seeing if this is for us.
1578 * These checks are supposed to be in an order that means easy
1579 * things are checked first to speed up processing.... however
1580 * this means that some packets will manage to get a long way
1581 * down this stack and then be rejected, but that's life.
1583 if ((ic
->type
!= ICMP_DEST_UNREACH
) &&
1584 (ic
->type
!= ICMP_SOURCE_QUENCH
) &&
1585 (ic
->type
!= ICMP_TIME_EXCEEDED
)) {
1590 /* Now find the contained IP header */
1591 offset
+= sizeof(_icmph
);
1592 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
1594 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1596 /* Special case for errors for IPIP packets */
1598 if (cih
->protocol
== IPPROTO_IPIP
) {
1599 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
)))
1601 /* Error for our IPIP must arrive at LOCAL_IN */
1602 if (!(skb_rtable(skb
)->rt_flags
& RTCF_LOCAL
))
1604 offset
+= cih
->ihl
* 4;
1605 cih
= skb_header_pointer(skb
, offset
, sizeof(_ciph
), &_ciph
);
1607 return NF_ACCEPT
; /* The packet looks wrong, ignore */
1611 pd
= ip_vs_proto_data_get(ipvs
, cih
->protocol
);
1616 /* Is the embedded protocol header present? */
1617 if (unlikely(cih
->frag_off
& htons(IP_OFFSET
) &&
1621 IP_VS_DBG_PKT(11, AF_INET
, pp
, skb
, offset
,
1622 "Checking incoming ICMP for");
1625 ip_vs_fill_iph_skb_icmp(AF_INET
, skb
, offset
, !ipip
, &ciph
);
1628 /* The embedded headers contain source and dest in reverse order.
1629 * For IPIP this is error for request, not for reply.
1631 cp
= pp
->conn_in_get(ipvs
, AF_INET
, skb
, &ciph
);
1636 if (!sysctl_schedule_icmp(ipvs
))
1639 if (!ip_vs_try_to_schedule(ipvs
, AF_INET
, skb
, pd
, &v
, &cp
, &ciph
))
1646 /* Ensure the checksum is correct */
1647 if (!skb_csum_unnecessary(skb
) && ip_vs_checksum_complete(skb
, ihl
)) {
1648 /* Failed checksum! */
1649 IP_VS_DBG(1, "Incoming ICMP: failed checksum from %pI4!\n",
1655 __be32 info
= ic
->un
.gateway
;
1656 __u8 type
= ic
->type
;
1657 __u8 code
= ic
->code
;
1659 /* Update the MTU */
1660 if (ic
->type
== ICMP_DEST_UNREACH
&&
1661 ic
->code
== ICMP_FRAG_NEEDED
) {
1662 struct ip_vs_dest
*dest
= cp
->dest
;
1663 u32 mtu
= ntohs(ic
->un
.frag
.mtu
);
1664 __be16 frag_off
= cih
->frag_off
;
1666 /* Strip outer IP and ICMP, go to IPIP header */
1667 if (pskb_pull(skb
, ihl
+ sizeof(_icmph
)) == NULL
)
1669 offset2
-= ihl
+ sizeof(_icmph
);
1670 skb_reset_network_header(skb
);
1671 IP_VS_DBG(12, "ICMP for IPIP %pI4->%pI4: mtu=%u\n",
1672 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
, mtu
);
1673 ipv4_update_pmtu(skb
, ipvs
->net
,
1675 /* Client uses PMTUD? */
1676 if (!(frag_off
& htons(IP_DF
)))
1678 /* Prefer the resulting PMTU */
1680 struct ip_vs_dest_dst
*dest_dst
;
1683 dest_dst
= rcu_dereference(dest
->dest_dst
);
1685 mtu
= dst_mtu(dest_dst
->dst_cache
);
1688 if (mtu
> 68 + sizeof(struct iphdr
))
1689 mtu
-= sizeof(struct iphdr
);
1692 /* Strip outer IP, ICMP and IPIP, go to IP header of
1695 if (pskb_pull(skb
, offset2
) == NULL
)
1697 skb_reset_network_header(skb
);
1698 IP_VS_DBG(12, "Sending ICMP for %pI4->%pI4: t=%u, c=%u, i=%u\n",
1699 &ip_hdr(skb
)->saddr
, &ip_hdr(skb
)->daddr
,
1700 type
, code
, ntohl(info
));
1701 icmp_send(skb
, type
, code
, info
);
1702 /* ICMP can be shorter but anyways, account it */
1703 ip_vs_out_stats(cp
, skb
);
1707 verdict
= NF_STOLEN
;
1711 /* do the statistics and put it back */
1712 ip_vs_in_stats(cp
, skb
);
1713 if (IPPROTO_TCP
== cih
->protocol
|| IPPROTO_UDP
== cih
->protocol
||
1714 IPPROTO_SCTP
== cih
->protocol
)
1715 offset
+= 2 * sizeof(__u16
);
1716 verdict
= ip_vs_icmp_xmit(skb
, cp
, pp
, offset
, hooknum
, &ciph
);
1719 if (likely(!new_cp
))
1720 __ip_vs_conn_put(cp
);
1727 #ifdef CONFIG_IP_VS_IPV6
1728 static int ip_vs_in_icmp_v6(struct netns_ipvs
*ipvs
, struct sk_buff
*skb
,
1729 int *related
, unsigned int hooknum
,
1730 struct ip_vs_iphdr
*iph
)
1732 struct icmp6hdr _icmph
, *ic
;
1733 struct ip_vs_iphdr ciph
= {.flags
= 0, .fragoffs
= 0};/*Contained IP */
1734 struct ip_vs_conn
*cp
;
1735 struct ip_vs_protocol
*pp
;
1736 struct ip_vs_proto_data
*pd
;
1737 unsigned int offset
, verdict
;
1738 bool new_cp
= false;
1742 ic
= frag_safe_skb_hp(skb
, iph
->len
, sizeof(_icmph
), &_icmph
, iph
);
1747 * Work through seeing if this is for us.
1748 * These checks are supposed to be in an order that means easy
1749 * things are checked first to speed up processing.... however
1750 * this means that some packets will manage to get a long way
1751 * down this stack and then be rejected, but that's life.
1753 if (ic
->icmp6_type
& ICMPV6_INFOMSG_MASK
) {
1757 /* Fragment header that is before ICMP header tells us that:
1758 * it's not an error message since they can't be fragmented.
1760 if (iph
->flags
& IP6_FH_F_FRAG
)
1763 IP_VS_DBG(8, "Incoming ICMPv6 (%d,%d) %pI6c->%pI6c\n",
1764 ic
->icmp6_type
, ntohs(icmpv6_id(ic
)),
1765 &iph
->saddr
, &iph
->daddr
);
1767 offset
= iph
->len
+ sizeof(_icmph
);
1768 if (!ip_vs_fill_iph_skb_icmp(AF_INET6
, skb
, offset
, true, &ciph
))
1771 pd
= ip_vs_proto_data_get(ipvs
, ciph
.protocol
);
1776 /* Cannot handle fragmented embedded protocol */
1780 IP_VS_DBG_PKT(11, AF_INET6
, pp
, skb
, offset
,
1781 "Checking incoming ICMPv6 for");
1783 /* The embedded headers contain source and dest in reverse order
1784 * if not from localhost
1786 cp
= pp
->conn_in_get(ipvs
, AF_INET6
, skb
, &ciph
);
1791 if (!sysctl_schedule_icmp(ipvs
))
1794 if (!ip_vs_try_to_schedule(ipvs
, AF_INET6
, skb
, pd
, &v
, &cp
, &ciph
))
1800 /* VS/TUN, VS/DR and LOCALNODE just let it go */
1801 if ((hooknum
== NF_INET_LOCAL_OUT
) &&
1802 (IP_VS_FWD_METHOD(cp
) != IP_VS_CONN_F_MASQ
)) {
1803 verdict
= NF_ACCEPT
;
1807 /* do the statistics and put it back */
1808 ip_vs_in_stats(cp
, skb
);
1810 /* Need to mangle contained IPv6 header in ICMPv6 packet */
1812 if (IPPROTO_TCP
== ciph
.protocol
|| IPPROTO_UDP
== ciph
.protocol
||
1813 IPPROTO_SCTP
== ciph
.protocol
)
1814 offset
+= 2 * sizeof(__u16
); /* Also mangle ports */
1816 verdict
= ip_vs_icmp_xmit_v6(skb
, cp
, pp
, offset
, hooknum
, &ciph
);
1819 if (likely(!new_cp
))
1820 __ip_vs_conn_put(cp
);
1830 * Check if it's for virtual services, look it up,
1831 * and send it on its way...
1834 ip_vs_in(struct netns_ipvs
*ipvs
, unsigned int hooknum
, struct sk_buff
*skb
, int af
)
1836 struct ip_vs_iphdr iph
;
1837 struct ip_vs_protocol
*pp
;
1838 struct ip_vs_proto_data
*pd
;
1839 struct ip_vs_conn
*cp
;
1841 int conn_reuse_mode
;
1844 /* Already marked as IPVS request or reply? */
1845 if (skb
->ipvs_property
)
1850 * - remote client: only PACKET_HOST
1851 * - route: used for struct net when skb->dev is unset
1853 if (unlikely((skb
->pkt_type
!= PACKET_HOST
&&
1854 hooknum
!= NF_INET_LOCAL_OUT
) ||
1856 ip_vs_fill_iph_skb(af
, skb
, false, &iph
);
1857 IP_VS_DBG_BUF(12, "packet type=%d proto=%d daddr=%s"
1858 " ignored in hook %u\n",
1859 skb
->pkt_type
, iph
.protocol
,
1860 IP_VS_DBG_ADDR(af
, &iph
.daddr
), hooknum
);
1863 /* ipvs enabled in this netns ? */
1864 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
1867 ip_vs_fill_iph_skb(af
, skb
, false, &iph
);
1869 /* Bad... Do not break raw sockets */
1870 sk
= skb_to_full_sk(skb
);
1871 if (unlikely(sk
&& hooknum
== NF_INET_LOCAL_OUT
&&
1874 if (sk
->sk_family
== PF_INET
&& inet_sk(sk
)->nodefrag
)
1878 #ifdef CONFIG_IP_VS_IPV6
1879 if (af
== AF_INET6
) {
1880 if (unlikely(iph
.protocol
== IPPROTO_ICMPV6
)) {
1882 int verdict
= ip_vs_in_icmp_v6(ipvs
, skb
, &related
,
1890 if (unlikely(iph
.protocol
== IPPROTO_ICMP
)) {
1892 int verdict
= ip_vs_in_icmp(ipvs
, skb
, &related
,
1899 /* Protocol supported? */
1900 pd
= ip_vs_proto_data_get(ipvs
, iph
.protocol
);
1901 if (unlikely(!pd
)) {
1902 /* The only way we'll see this packet again is if it's
1903 * encapsulated, so mark it with ipvs_property=1 so we
1904 * skip it if we're ignoring tunneled packets
1906 if (sysctl_ignore_tunneled(ipvs
))
1907 skb
->ipvs_property
= 1;
1913 * Check if the packet belongs to an existing connection entry
1915 cp
= pp
->conn_in_get(ipvs
, af
, skb
, &iph
);
1917 conn_reuse_mode
= sysctl_conn_reuse_mode(ipvs
);
1918 if (conn_reuse_mode
&& !iph
.fragoffs
&& is_new_conn(skb
, &iph
) && cp
) {
1919 bool uses_ct
= false, resched
= false;
1921 if (unlikely(sysctl_expire_nodest_conn(ipvs
)) && cp
->dest
&&
1922 unlikely(!atomic_read(&cp
->dest
->weight
))) {
1924 uses_ct
= ip_vs_conn_uses_conntrack(cp
, skb
);
1925 } else if (is_new_conn_expected(cp
, conn_reuse_mode
)) {
1926 uses_ct
= ip_vs_conn_uses_conntrack(cp
, skb
);
1927 if (!atomic_read(&cp
->n_control
)) {
1930 /* Do not reschedule controlling connection
1931 * that uses conntrack while it is still
1932 * referenced by controlled connection(s).
1939 if (!atomic_read(&cp
->n_control
))
1940 ip_vs_conn_expire_now(cp
);
1941 __ip_vs_conn_put(cp
);
1948 if (unlikely(!cp
)) {
1951 if (!ip_vs_try_to_schedule(ipvs
, af
, skb
, pd
, &v
, &cp
, &iph
))
1955 IP_VS_DBG_PKT(11, af
, pp
, skb
, iph
.off
, "Incoming packet");
1957 /* Check the server status */
1958 if (cp
->dest
&& !(cp
->dest
->flags
& IP_VS_DEST_F_AVAILABLE
)) {
1959 /* the destination server is not available */
1961 if (sysctl_expire_nodest_conn(ipvs
)) {
1962 /* try to expire the connection immediately */
1963 ip_vs_conn_expire_now(cp
);
1965 /* don't restart its timer, and silently
1967 __ip_vs_conn_put(cp
);
1971 ip_vs_in_stats(cp
, skb
);
1972 ip_vs_set_state(cp
, IP_VS_DIR_INPUT
, skb
, pd
);
1973 if (cp
->packet_xmit
)
1974 ret
= cp
->packet_xmit(skb
, cp
, pp
, &iph
);
1975 /* do not touch skb anymore */
1977 IP_VS_DBG_RL("warning: packet_xmit is null");
1981 /* Increase its packet counter and check if it is needed
1982 * to be synchronized
1984 * Sync connection if it is about to close to
1985 * encorage the standby servers to update the connections timeout
1987 * For ONE_PKT let ip_vs_sync_conn() do the filter work.
1990 if (cp
->flags
& IP_VS_CONN_F_ONE_PACKET
)
1991 pkts
= sysctl_sync_threshold(ipvs
);
1993 pkts
= atomic_add_return(1, &cp
->in_pkts
);
1995 if (ipvs
->sync_state
& IP_VS_STATE_MASTER
)
1996 ip_vs_sync_conn(ipvs
, cp
, pkts
);
1997 else if ((cp
->flags
& IP_VS_CONN_F_ONE_PACKET
) && cp
->control
)
1998 /* increment is done inside ip_vs_sync_conn too */
1999 atomic_inc(&cp
->control
->in_pkts
);
2006 * AF_INET handler in NF_INET_LOCAL_IN chain
2007 * Schedule and forward packets from remote clients
2010 ip_vs_remote_request4(void *priv
, struct sk_buff
*skb
,
2011 const struct nf_hook_state
*state
)
2013 return ip_vs_in(net_ipvs(state
->net
), state
->hook
, skb
, AF_INET
);
2017 * AF_INET handler in NF_INET_LOCAL_OUT chain
2018 * Schedule and forward packets from local clients
2021 ip_vs_local_request4(void *priv
, struct sk_buff
*skb
,
2022 const struct nf_hook_state
*state
)
2024 return ip_vs_in(net_ipvs(state
->net
), state
->hook
, skb
, AF_INET
);
2027 #ifdef CONFIG_IP_VS_IPV6
2030 * AF_INET6 handler in NF_INET_LOCAL_IN chain
2031 * Schedule and forward packets from remote clients
2034 ip_vs_remote_request6(void *priv
, struct sk_buff
*skb
,
2035 const struct nf_hook_state
*state
)
2037 return ip_vs_in(net_ipvs(state
->net
), state
->hook
, skb
, AF_INET6
);
2041 * AF_INET6 handler in NF_INET_LOCAL_OUT chain
2042 * Schedule and forward packets from local clients
2045 ip_vs_local_request6(void *priv
, struct sk_buff
*skb
,
2046 const struct nf_hook_state
*state
)
2048 return ip_vs_in(net_ipvs(state
->net
), state
->hook
, skb
, AF_INET6
);
2055 * It is hooked at the NF_INET_FORWARD chain, in order to catch ICMP
2056 * related packets destined for 0.0.0.0/0.
2057 * When fwmark-based virtual service is used, such as transparent
2058 * cache cluster, TCP packets can be marked and routed to ip_vs_in,
2059 * but ICMP destined for 0.0.0.0/0 cannot not be easily marked and
2060 * sent to ip_vs_in_icmp. So, catch them at the NF_INET_FORWARD chain
2061 * and send them to ip_vs_in_icmp.
2064 ip_vs_forward_icmp(void *priv
, struct sk_buff
*skb
,
2065 const struct nf_hook_state
*state
)
2068 struct netns_ipvs
*ipvs
= net_ipvs(state
->net
);
2070 if (ip_hdr(skb
)->protocol
!= IPPROTO_ICMP
)
2073 /* ipvs enabled in this netns ? */
2074 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
2077 return ip_vs_in_icmp(ipvs
, skb
, &r
, state
->hook
);
2080 #ifdef CONFIG_IP_VS_IPV6
2082 ip_vs_forward_icmp_v6(void *priv
, struct sk_buff
*skb
,
2083 const struct nf_hook_state
*state
)
2086 struct netns_ipvs
*ipvs
= net_ipvs(state
->net
);
2087 struct ip_vs_iphdr iphdr
;
2089 ip_vs_fill_iph_skb(AF_INET6
, skb
, false, &iphdr
);
2090 if (iphdr
.protocol
!= IPPROTO_ICMPV6
)
2093 /* ipvs enabled in this netns ? */
2094 if (unlikely(sysctl_backup_only(ipvs
) || !ipvs
->enable
))
2097 return ip_vs_in_icmp_v6(ipvs
, skb
, &r
, state
->hook
, &iphdr
);
2102 static struct nf_hook_ops ip_vs_ops
[] __read_mostly
= {
2103 /* After packet filtering, change source only for VS/NAT */
2105 .hook
= ip_vs_reply4
,
2107 .hooknum
= NF_INET_LOCAL_IN
,
2108 .priority
= NF_IP_PRI_NAT_SRC
- 2,
2110 /* After packet filtering, forward packet through VS/DR, VS/TUN,
2111 * or VS/NAT(change destination), so that filtering rules can be
2112 * applied to IPVS. */
2114 .hook
= ip_vs_remote_request4
,
2116 .hooknum
= NF_INET_LOCAL_IN
,
2117 .priority
= NF_IP_PRI_NAT_SRC
- 1,
2119 /* Before ip_vs_in, change source only for VS/NAT */
2121 .hook
= ip_vs_local_reply4
,
2123 .hooknum
= NF_INET_LOCAL_OUT
,
2124 .priority
= NF_IP_PRI_NAT_DST
+ 1,
2126 /* After mangle, schedule and forward local requests */
2128 .hook
= ip_vs_local_request4
,
2130 .hooknum
= NF_INET_LOCAL_OUT
,
2131 .priority
= NF_IP_PRI_NAT_DST
+ 2,
2133 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2134 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2136 .hook
= ip_vs_forward_icmp
,
2138 .hooknum
= NF_INET_FORWARD
,
2141 /* After packet filtering, change source only for VS/NAT */
2143 .hook
= ip_vs_reply4
,
2145 .hooknum
= NF_INET_FORWARD
,
2148 #ifdef CONFIG_IP_VS_IPV6
2149 /* After packet filtering, change source only for VS/NAT */
2151 .hook
= ip_vs_reply6
,
2153 .hooknum
= NF_INET_LOCAL_IN
,
2154 .priority
= NF_IP6_PRI_NAT_SRC
- 2,
2156 /* After packet filtering, forward packet through VS/DR, VS/TUN,
2157 * or VS/NAT(change destination), so that filtering rules can be
2158 * applied to IPVS. */
2160 .hook
= ip_vs_remote_request6
,
2162 .hooknum
= NF_INET_LOCAL_IN
,
2163 .priority
= NF_IP6_PRI_NAT_SRC
- 1,
2165 /* Before ip_vs_in, change source only for VS/NAT */
2167 .hook
= ip_vs_local_reply6
,
2169 .hooknum
= NF_INET_LOCAL_OUT
,
2170 .priority
= NF_IP6_PRI_NAT_DST
+ 1,
2172 /* After mangle, schedule and forward local requests */
2174 .hook
= ip_vs_local_request6
,
2176 .hooknum
= NF_INET_LOCAL_OUT
,
2177 .priority
= NF_IP6_PRI_NAT_DST
+ 2,
2179 /* After packet filtering (but before ip_vs_out_icmp), catch icmp
2180 * destined for 0.0.0.0/0, which is for incoming IPVS connections */
2182 .hook
= ip_vs_forward_icmp_v6
,
2184 .hooknum
= NF_INET_FORWARD
,
2187 /* After packet filtering, change source only for VS/NAT */
2189 .hook
= ip_vs_reply6
,
2191 .hooknum
= NF_INET_FORWARD
,
2197 * Initialize IP Virtual Server netns mem.
2199 static int __net_init
__ip_vs_init(struct net
*net
)
2201 struct netns_ipvs
*ipvs
;
2203 ipvs
= net_generic(net
, ip_vs_net_id
);
2207 /* Hold the beast until a service is registerd */
2210 /* Counters used for creating unique names */
2211 ipvs
->gen
= atomic_read(&ipvs_netns_cnt
);
2212 atomic_inc(&ipvs_netns_cnt
);
2215 if (ip_vs_estimator_net_init(ipvs
) < 0)
2216 goto estimator_fail
;
2218 if (ip_vs_control_net_init(ipvs
) < 0)
2221 if (ip_vs_protocol_net_init(ipvs
) < 0)
2224 if (ip_vs_app_net_init(ipvs
) < 0)
2227 if (ip_vs_conn_net_init(ipvs
) < 0)
2230 if (ip_vs_sync_net_init(ipvs
) < 0)
2233 printk(KERN_INFO
"IPVS: Creating netns size=%zu id=%d\n",
2234 sizeof(struct netns_ipvs
), ipvs
->gen
);
2241 ip_vs_conn_net_cleanup(ipvs
);
2243 ip_vs_app_net_cleanup(ipvs
);
2245 ip_vs_protocol_net_cleanup(ipvs
);
2247 ip_vs_control_net_cleanup(ipvs
);
2249 ip_vs_estimator_net_cleanup(ipvs
);
2255 static void __net_exit
__ip_vs_cleanup(struct net
*net
)
2257 struct netns_ipvs
*ipvs
= net_ipvs(net
);
2259 ip_vs_service_net_cleanup(ipvs
); /* ip_vs_flush() with locks */
2260 ip_vs_conn_net_cleanup(ipvs
);
2261 ip_vs_app_net_cleanup(ipvs
);
2262 ip_vs_protocol_net_cleanup(ipvs
);
2263 ip_vs_control_net_cleanup(ipvs
);
2264 ip_vs_estimator_net_cleanup(ipvs
);
2265 IP_VS_DBG(2, "ipvs netns %d released\n", ipvs
->gen
);
2269 static void __net_exit
__ip_vs_dev_cleanup(struct net
*net
)
2271 struct netns_ipvs
*ipvs
= net_ipvs(net
);
2273 ipvs
->enable
= 0; /* Disable packet reception */
2275 ip_vs_sync_net_cleanup(ipvs
);
2279 static struct pernet_operations ipvs_core_ops
= {
2280 .init
= __ip_vs_init
,
2281 .exit
= __ip_vs_cleanup
,
2282 .id
= &ip_vs_net_id
,
2283 .size
= sizeof(struct netns_ipvs
),
2286 static struct pernet_operations ipvs_core_dev_ops
= {
2287 .exit
= __ip_vs_dev_cleanup
,
2291 * Initialize IP Virtual Server
2293 static int __init
ip_vs_init(void)
2297 ret
= ip_vs_control_init();
2299 pr_err("can't setup control.\n");
2303 ip_vs_protocol_init();
2305 ret
= ip_vs_conn_init();
2307 pr_err("can't setup connection table.\n");
2308 goto cleanup_protocol
;
2311 ret
= register_pernet_subsys(&ipvs_core_ops
); /* Alloc ip_vs struct */
2315 ret
= register_pernet_device(&ipvs_core_dev_ops
);
2319 ret
= nf_register_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2321 pr_err("can't register hooks.\n");
2325 ret
= ip_vs_register_nl_ioctl();
2327 pr_err("can't register netlink/ioctl.\n");
2331 pr_info("ipvs loaded.\n");
2336 nf_unregister_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2338 unregister_pernet_device(&ipvs_core_dev_ops
);
2340 unregister_pernet_subsys(&ipvs_core_ops
);
2342 ip_vs_conn_cleanup();
2344 ip_vs_protocol_cleanup();
2345 ip_vs_control_cleanup();
2350 static void __exit
ip_vs_cleanup(void)
2352 ip_vs_unregister_nl_ioctl();
2353 nf_unregister_hooks(ip_vs_ops
, ARRAY_SIZE(ip_vs_ops
));
2354 unregister_pernet_device(&ipvs_core_dev_ops
);
2355 unregister_pernet_subsys(&ipvs_core_ops
); /* free ip_vs struct */
2356 ip_vs_conn_cleanup();
2357 ip_vs_protocol_cleanup();
2358 ip_vs_control_cleanup();
2359 pr_info("ipvs unloaded.\n");
2362 module_init(ip_vs_init
);
2363 module_exit(ip_vs_cleanup
);
2364 MODULE_LICENSE("GPL");